Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A system configured to generate a high resolution image for a specimen from a low resolution image of the specimen, comprising: one or more computer subsystems configured for acquiring a low resolution image of a specimen; and one or more components executed by the one or more computer subsystems, wherein the one or more components comprise: a deep convolutional neural network, wherein the deep convolutional neural network comprises: one or more first layers configured for generating a representation of the low resolution image; and one or more second layers configured for generating a high resolution image for the specimen from the representation of the low resolution image, wherein the one or more second layers comprise a final layer configured to output the high resolution image, and wherein the final layer is further configured as a sub-pixel convolution layer.
The system enhances low-resolution images of specimens into high-resolution versions using deep learning. In microscopy and imaging applications, low-resolution images often lack sufficient detail for accurate analysis, limiting scientific and medical research. The system addresses this by employing a deep convolutional neural network (CNN) to upscale images while preserving critical features. The system includes computer subsystems that acquire low-resolution images of specimens. A deep CNN processes these images through multiple layers. The first layers extract and encode key features from the low-resolution input, creating a compact representation. The subsequent layers reconstruct a high-resolution image from this representation. The final layer, a sub-pixel convolution layer, refines the output by adjusting pixel positions and interpolating finer details, improving resolution without artifacts. This approach leverages sub-pixel convolution to enhance image quality, which is particularly useful in fields like pathology, where high-resolution imaging is essential for diagnosis. The system automates the upscaling process, reducing reliance on manual adjustments or traditional interpolation methods that may introduce distortions. The deep learning model is trained to recognize and reconstruct fine details, making it suitable for various specimen types.
2. The system of claim 1 , wherein the deep convolutional neural network is configured such that the high resolution image generated by the one or more second layers has less noise than the low resolution image.
This invention relates to image processing systems that use deep convolutional neural networks (CNNs) to enhance image resolution while reducing noise. The system addresses the challenge of generating high-resolution images from low-resolution inputs, where traditional methods often introduce artifacts or fail to effectively suppress noise. The core innovation involves a multi-layered CNN architecture where one or more second layers are specifically designed to produce a high-resolution output image with reduced noise compared to the original low-resolution input. The network is trained to learn the mapping between low-resolution and high-resolution images, leveraging convolutional operations to capture spatial hierarchies and patterns. The second layers refine the output by applying noise reduction techniques, such as denoising filters or adversarial training, ensuring the final image retains fine details while minimizing distortions. This approach improves upon prior art by integrating noise suppression directly into the resolution enhancement process, rather than treating it as a separate post-processing step. The system is applicable in medical imaging, satellite imagery, and surveillance, where high-quality images are critical. The invention ensures that the generated high-resolution images are both sharp and free from the noise present in the original low-resolution input.
3. The system of claim 1 , wherein the deep convolutional neural network is configured such that the high resolution image generated by the one or more second layers retains structural and spatial features of the low resolution image.
This invention relates to image processing systems that enhance low-resolution images using deep convolutional neural networks (CNNs). The problem addressed is the loss of structural and spatial details when upscaling low-resolution images, which traditional methods often fail to preserve accurately. The system includes a CNN with multiple layers, where initial layers process the low-resolution input image, and subsequent layers generate a high-resolution output. The key innovation lies in the network's configuration, ensuring that the high-resolution image retains the original structural and spatial features of the low-resolution input. This is achieved through specialized layers that maintain geometric and texture consistency during upscaling. The system may also incorporate additional processing steps, such as noise reduction or feature extraction, to further refine the output. By leveraging deep learning techniques, the system improves upon conventional interpolation methods, which often produce blurry or distorted results. The approach is particularly useful in applications requiring high-fidelity image reconstruction, such as medical imaging, satellite imagery, and digital photography. The network's architecture is designed to learn and preserve intricate details, making it adaptable to various imaging scenarios while maintaining computational efficiency.
4. The system of claim 1 , wherein the one or more components further comprise a context aware loss module configured to train the deep convolutional neural network, wherein during training of the deep convolutional neural network, the one or more computer subsystems input the high resolution image generated by the one or more second layers and a corresponding, known high resolution image for the specimen into the context aware loss module and the context aware loss module determines context aware loss in the high resolution image generated by the one or more second layers compared to the corresponding, known high resolution image.
This invention relates to a system for high-resolution image reconstruction using deep convolutional neural networks (CNNs), particularly for enhancing the resolution of images of specimens. The system addresses the challenge of accurately reconstructing high-resolution images from lower-resolution inputs, which is critical in fields like microscopy, medical imaging, and remote sensing where high-resolution details are essential for analysis. The system includes a deep CNN with multiple layers, where initial layers process the input image and subsequent layers generate a high-resolution output. A key component is a context-aware loss module, which improves training by comparing the generated high-resolution image to a known high-resolution reference image of the same specimen. During training, the module calculates a context-aware loss, which quantifies discrepancies between the generated and reference images, guiding the CNN to refine its output. This approach ensures that the reconstructed image retains fine details and structural accuracy, addressing limitations of traditional loss functions that may overlook contextual relationships in the data. The system is designed to enhance image reconstruction quality by leveraging contextual information during training, making it particularly useful in applications requiring precise high-resolution imaging.
5. The system of claim 4 , wherein the context aware loss comprises content loss, style loss, and total variation regularization.
A system for image processing or generation that incorporates context-aware loss functions to improve output quality. The system addresses the challenge of producing high-fidelity images by combining multiple loss components to guide the optimization process. The context-aware loss includes content loss, which ensures the generated image retains key structural features from a reference image; style loss, which preserves stylistic elements such as texture and color patterns; and total variation regularization, which smooths the output while maintaining sharp edges. These components work together to balance fidelity, aesthetics, and computational efficiency. The system is particularly useful in applications like image synthesis, style transfer, and super-resolution, where maintaining both semantic and perceptual quality is critical. By dynamically adjusting the weights of these loss terms, the system adapts to different input conditions and desired output characteristics, resulting in more natural and visually coherent results. The approach improves over traditional methods that rely on single loss functions or fixed combinations, offering greater flexibility and performance in diverse scenarios.
6. The system of claim 5 , wherein the content loss comprises loss in low level features of the corresponding, known high resolution image.
The invention relates to image processing systems designed to address the problem of content loss in low-level features during image resolution enhancement. The system is configured to analyze and restore fine details in high-resolution images that may be degraded or lost during processing. Specifically, it focuses on preserving or reconstructing low-level features such as edges, textures, and fine structures that are critical for perceptual quality but often diminished in standard upscaling techniques. The system operates by comparing a processed image to a known high-resolution reference image to identify discrepancies in low-level features. It then applies corrective measures to mitigate these losses, ensuring that the output image retains the same level of detail and fidelity as the original high-resolution image. This approach is particularly useful in applications where image quality is paramount, such as medical imaging, satellite imagery, and high-definition video production. The system may incorporate machine learning models or traditional image processing algorithms to detect and compensate for feature loss. It can be integrated into existing image processing pipelines or deployed as a standalone tool for post-processing. The primary goal is to enhance image quality by minimizing distortions and artifacts that arise from resolution scaling or compression, thereby improving visual accuracy and user experience.
7. The system of claim 5 , wherein the style loss comprises loss in one or more abstract entities that qualitatively define the corresponding, known high resolution image.
The invention relates to image processing systems, specifically those designed to preserve stylistic features during image transformation or enhancement. The problem addressed is the loss of abstract stylistic elements when converting or generating images, particularly when working with high-resolution images. These abstract entities may include texture patterns, color distributions, or other qualitative characteristics that define the visual style of an image. The system includes a neural network or similar processing module that analyzes a known high-resolution image to identify and extract these abstract stylistic features. During image transformation, such as super-resolution or style transfer, the system calculates a "style loss" metric that quantifies deviations from these extracted features. This loss metric is used to guide the transformation process, ensuring that the output image retains the original stylistic qualities. The system may employ techniques like feature extraction from intermediate layers of a convolutional neural network to capture these abstract entities. The invention improves upon prior methods by focusing on qualitative, high-level stylistic features rather than low-level pixel-based metrics. This approach ensures that the transformed image maintains the intended artistic or perceptual qualities, which is particularly important in applications like digital art, medical imaging, or satellite imagery where stylistic consistency is critical. The system can be integrated into existing image processing pipelines to enhance their ability to preserve stylistic integrity during transformations.
8. The system of claim 4 , wherein the context aware loss module comprises a pre-trained VGG network.
A system for context-aware data processing leverages a pre-trained VGG network within a context-aware loss module to enhance accuracy in tasks such as image recognition or classification. The VGG network, a deep convolutional neural network known for its effectiveness in feature extraction, is integrated to analyze input data and generate context-aware loss values. These values are used to refine the system's output by adjusting predictions based on contextual information derived from the network's learned features. The system addresses the challenge of improving model performance by incorporating contextual understanding, which traditional loss functions often overlook. By utilizing a pre-trained VGG network, the system benefits from transfer learning, reducing the need for extensive training data while maintaining high accuracy. The context-aware loss module dynamically adapts to variations in input data, ensuring robust performance across diverse scenarios. This approach is particularly valuable in applications where contextual nuances significantly impact results, such as medical imaging, autonomous driving, or natural language processing. The system's architecture ensures scalability and adaptability, making it suitable for integration into various machine learning pipelines.
9. The system of claim 4 , wherein the one or more components further comprise a tuning module configured to determine one or more parameters of the deep convolutional neural network based on the context aware loss.
A system for optimizing deep convolutional neural networks (CNNs) in machine learning applications addresses the challenge of improving model performance by dynamically adjusting network parameters based on contextual data. The system includes a context-aware loss function that evaluates model outputs against contextual factors such as input data characteristics, environmental conditions, or task-specific requirements. This loss function generates a context-aware loss value that reflects the model's performance in varying scenarios. A tuning module uses this loss value to adjust one or more parameters of the CNN, such as layer weights, activation functions, or network architecture, to enhance accuracy, efficiency, or adaptability. The system may also incorporate a data preprocessing module to prepare input data for the CNN and an output analysis module to assess model predictions. By dynamically tuning the CNN based on contextual feedback, the system improves model generalization and robustness across diverse operating conditions. This approach is particularly useful in applications where input data variability or environmental factors significantly impact model performance, such as autonomous systems, real-time decision-making, or adaptive control systems. The tuning module ensures the CNN remains optimized for its specific use case, reducing the need for manual parameter adjustments and improving overall system reliability.
10. The system of claim 1 , wherein the one or more computer subsystems are further configured to perform one or more metrology measurements for the specimen based on the high resolution image generated by the one or more second layers.
11. The system of claim 1 , wherein the deep convolutional neural network functions independently of an imaging system that generated the low resolution image.
This invention relates to image processing systems that enhance low-resolution images using deep convolutional neural networks (CNNs). The core problem addressed is the need to improve image quality without relying on the original imaging system that captured the low-resolution input. Traditional super-resolution techniques often depend on characteristics of the imaging hardware, such as sensor noise or lens distortions, to reconstruct higher-resolution details. However, this approach limits flexibility when the imaging system is unknown or unavailable. The system employs a deep CNN trained to generate high-resolution images from low-resolution inputs independently of the original imaging system. The neural network is designed to learn generic image features and patterns that can be applied universally, regardless of the source of the low-resolution image. This independence allows the system to function effectively with images from diverse devices, including cameras, scanners, or other imaging systems, without requiring prior knowledge of their specifications or calibration data. The CNN architecture may include multiple convolutional layers to extract hierarchical features, followed by upsampling layers to produce the enhanced output. The training process likely involves large datasets of low-resolution and corresponding high-resolution image pairs to ensure robust performance across different scenarios. This approach enables high-quality image reconstruction in applications where the original imaging system is not accessible or its parameters are unknown.
12. The system of claim 1 , wherein the low resolution image is generated by one imaging system having a first imaging platform, wherein the one or more computer subsystems are further configured for acquiring another low resolution image generated for another specimen by another imaging system having a second imaging platform that is different than the first imaging platform, wherein the one or more first layers are configured for generating a representation of the another low resolution image, and wherein the one or more second layers are further configured for generating a high resolution image for the another specimen from the representation of the another low resolution image.
The invention relates to a system for generating high-resolution images from low-resolution images acquired by different imaging platforms. The system addresses the challenge of producing high-quality images when multiple imaging systems with varying resolutions and platforms are used, ensuring consistency and accuracy across different specimens. The system includes one or more computer subsystems configured to process low-resolution images from multiple imaging systems. Each imaging system operates on a distinct imaging platform, meaning they may have different hardware, sensors, or configurations. The system acquires a low-resolution image from a first imaging platform and generates a representation of this image using one or more first layers, which likely involve preprocessing or feature extraction. The system then uses one or more second layers to reconstruct a high-resolution image from this representation. Additionally, the system can process another low-resolution image from a second, different imaging platform. The first layers generate a representation of this second low-resolution image, and the second layers produce a high-resolution image for the corresponding specimen. This ensures that images from different imaging systems can be standardized and enhanced to a consistent high resolution, improving analysis and comparison across diverse imaging sources. The system thus enables seamless integration and high-quality output from heterogeneous imaging platforms.
13. The system of claim 12 , wherein the first imaging platform is an electron beam imaging platform, and wherein the second imaging platform is an optical imaging platform.
This invention relates to a multi-modal imaging system designed to enhance the analysis of samples by combining electron beam imaging with optical imaging. The system addresses the limitations of single-modal imaging by integrating two distinct imaging platforms to provide complementary data. The first imaging platform is an electron beam imaging system, which offers high-resolution imaging at the nanoscale, revealing fine structural details of the sample. The second imaging platform is an optical imaging system, which provides broader field-of-view imaging and can capture dynamic processes or larger-scale features that are difficult to observe with electron microscopy alone. The system is configured to align and correlate the data from both platforms, allowing users to leverage the strengths of each modality. This dual-platform approach enables more comprehensive sample analysis, particularly in fields such as materials science, biology, and semiconductor inspection, where both high-resolution and contextual information are critical. The system may include mechanisms to ensure precise spatial registration between the two imaging modalities, ensuring accurate overlay of the data for analysis. The integration of electron beam and optical imaging in a single system improves workflow efficiency and reduces the need for separate, standalone instruments.
14. The system of claim 12 , wherein the first and second imaging platforms are different optical imaging platforms.
This invention relates to a system for capturing and processing images using multiple optical imaging platforms to address challenges in imaging accuracy, efficiency, or coverage. The system includes at least two distinct optical imaging platforms, each configured to capture images of a target area from different perspectives or using different imaging techniques. The platforms may include cameras, sensors, or other optical devices with varying capabilities, such as resolution, field of view, or spectral sensitivity. The system integrates data from these platforms to generate a composite or enhanced output, improving image quality, depth perception, or coverage of the target area. The different platforms may operate simultaneously or sequentially, and the system may synchronize their operations to ensure alignment or consistency in the captured data. This approach enables applications in fields such as surveillance, medical imaging, or industrial inspection, where multiple imaging perspectives or modalities are beneficial. The system may also include processing components to analyze, merge, or interpret the combined data from the platforms.
15. The system of claim 12 , wherein the first and second imaging platforms are different electron beam imaging platforms.
This invention relates to a system for imaging using multiple electron beam platforms. The system addresses the challenge of obtaining high-resolution, artifact-free images by leveraging distinct electron beam imaging platforms to capture complementary data. The first and second imaging platforms operate independently, each generating electron beam images of a sample from different perspectives or under different conditions. The system integrates these images to produce a composite output with enhanced accuracy and detail. The platforms may differ in their electron beam configurations, detection methods, or operational parameters, allowing for the capture of diverse imaging data that would be difficult or impossible to obtain with a single platform. This approach improves image quality by mitigating limitations inherent to individual electron beam systems, such as resolution constraints, contrast issues, or artifacts. The system is particularly useful in fields like materials science, semiconductor inspection, and biological imaging, where high-fidelity imaging is critical. By combining data from multiple electron beam sources, the invention enables more comprehensive and reliable analysis of samples.
16. The system of claim 1 , wherein the low resolution image is generated by an electron beam based imaging system.
This invention relates to an imaging system that generates a low-resolution image using an electron beam-based imaging system. The system includes a high-resolution imaging device that captures a high-resolution image of a sample, and a processing unit that generates a low-resolution image from the high-resolution image. The low-resolution image is then used to guide the electron beam-based imaging system to capture additional high-resolution images of specific regions of interest within the sample. The system also includes a display unit that presents the low-resolution image to a user, allowing them to select regions for further high-resolution imaging. The electron beam-based imaging system may include a scanning electron microscope (SEM) or a transmission electron microscope (TEM), which provides high-resolution imaging capabilities. The processing unit may apply downsampling, averaging, or other techniques to generate the low-resolution image from the high-resolution image. The system improves efficiency by reducing the time and computational resources required to analyze large samples by focusing only on regions of interest. This approach is particularly useful in fields such as materials science, semiconductor inspection, and biological imaging, where high-resolution imaging of specific areas is critical.
17. The system of claim 1 , wherein the low resolution image is generated by an optical based imaging system.
This invention relates to an imaging system that generates a low-resolution image using an optical-based imaging system. The system includes a sensor array configured to capture light from a scene and produce a low-resolution image. The sensor array may be a single-pixel sensor or a multi-pixel sensor with reduced resolution. The optical-based imaging system may include lenses, mirrors, or other optical components to focus or direct light onto the sensor array. The system may also include processing circuitry to process the captured image data, such as noise reduction, contrast enhancement, or other image processing techniques. The low-resolution image can be used for various applications, including surveillance, medical imaging, or industrial inspection, where high-resolution details are not required, or where computational efficiency is prioritized. The optical-based imaging system may be designed to operate in specific wavelength ranges, such as visible light, infrared, or ultraviolet, depending on the application. The system may also include calibration mechanisms to ensure accurate image capture and processing. The invention aims to provide a cost-effective and efficient imaging solution for scenarios where high-resolution imaging is unnecessary or impractical.
18. The system of claim 1 , wherein the low resolution image is generated by an inspection system.
The system relates to image processing in industrial inspection, specifically for generating and analyzing low-resolution images to detect defects or anomalies in manufactured products. The core problem addressed is the need for efficient, high-speed defect detection in production lines where high-resolution imaging may be impractical due to processing constraints or cost. The system generates a low-resolution image of an object under inspection, which is then processed to identify defects. The low-resolution image is produced by an inspection system, which may include cameras, sensors, or other imaging devices optimized for rapid capture and analysis. The system further includes a processing module that compares the low-resolution image against reference data or patterns to detect deviations indicative of defects. The processing module may apply machine learning algorithms, pattern recognition, or statistical analysis to enhance detection accuracy. The system may also include feedback mechanisms to adjust inspection parameters dynamically based on detected defects, improving overall quality control. The use of low-resolution imaging reduces computational overhead while maintaining sufficient detail for defect detection, making the system suitable for high-throughput manufacturing environments. The inspection system may be integrated into automated production lines or standalone quality control stations, ensuring real-time monitoring and defect identification.
19. The system of claim 1 , wherein the specimen is a wafer.
A system for handling and processing semiconductor wafers includes a wafer handling mechanism designed to securely grip and transport wafers during fabrication processes. The system ensures precise positioning and alignment of the wafer to prevent damage and maintain manufacturing accuracy. The wafer handling mechanism may incorporate vacuum-based or mechanical clamping systems to hold the wafer in place during operations such as etching, deposition, or inspection. The system may also include sensors to monitor wafer orientation, temperature, or other critical parameters to optimize processing conditions. Additionally, the system may feature automated control mechanisms to adjust handling parameters based on real-time feedback, ensuring consistent performance across multiple wafers. The wafer handling mechanism is integrated into a larger semiconductor fabrication setup, which may include robotic arms, conveyor systems, or other automated components to streamline wafer movement between processing stations. The system is designed to minimize contamination and reduce handling time, improving overall yield and efficiency in semiconductor manufacturing.
20. The system of claim 1 , wherein the specimen is a reticle.
A system for inspecting reticles in semiconductor manufacturing is disclosed. The system addresses the challenge of detecting defects in reticles, which are critical components used in photolithography to transfer circuit patterns onto semiconductor wafers. Defects in reticles can lead to faulty semiconductor devices, making accurate inspection essential. The system includes an imaging device configured to capture images of the reticle, a processing unit that analyzes the images to identify defects, and a database for storing inspection results. The imaging device may use optical or electron microscopy to achieve high-resolution imaging. The processing unit applies image processing algorithms, such as pattern recognition and defect detection, to compare the captured images against a reference or expected pattern. The system may also include a calibration module to ensure the imaging device maintains accuracy over time. Additionally, the system may support automated defect classification, allowing for rapid identification of different defect types, such as particles, scratches, or pattern distortions. The inspection results are stored in a database for traceability and quality control. The system may further integrate with manufacturing execution systems to trigger corrective actions when defects are detected. This ensures that only defect-free reticles are used in semiconductor production, improving yield and reliability.
21. The system of claim 1 , wherein the deep convolutional neural network outputs the high resolution image at a throughput that is higher than a throughput for generating the high resolution image with a high resolution imaging system.
The invention relates to a system for generating high-resolution images using a deep convolutional neural network (CNN) that achieves higher throughput compared to traditional high-resolution imaging systems. The system addresses the challenge of efficiently producing high-resolution images without the computational and hardware limitations of conventional imaging methods. The deep CNN processes input data to generate a high-resolution output image at a faster rate than traditional imaging systems, which often rely on slower, resource-intensive processes. The system leverages the neural network's ability to learn and apply complex patterns to reconstruct high-resolution details from lower-resolution or partially processed input data. This approach reduces the need for expensive, high-resolution imaging hardware and accelerates image generation, making it suitable for applications requiring real-time or high-volume image processing, such as medical imaging, satellite imaging, or industrial inspection. The neural network is trained to optimize both image quality and processing speed, ensuring that the output meets high-resolution standards while maintaining superior throughput. The system may also include preprocessing and postprocessing modules to enhance input data quality and refine the final output, further improving efficiency and accuracy.
22. A system configured to generate a high resolution image for a specimen from a low resolution image of the specimen, comprising: an imaging subsystem configured for generating a low resolution image of a specimen; one or more computer subsystems configured for acquiring the low resolution image of the specimen; and one or more components executed by the one or more computer subsystems, wherein the one or more components comprise: a deep convolutional neural network, wherein the deep convolutional neural network comprises: one or more first layers configured for generating a representation of the low resolution image; and one or more second layers configured for generating a high resolution image for the specimen from the representation of the low resolution image, wherein the one or more second layers comprise a final layer configured to output the high resolution image, and wherein the final layer is further configured as a sub-pixel convolution layer.
The system enhances low-resolution images of specimens to produce high-resolution outputs using deep learning techniques. In microscopy and imaging applications, low-resolution images often lack sufficient detail for accurate analysis, limiting scientific and medical research. The system addresses this by employing a deep convolutional neural network (CNN) to upscale images while preserving critical features. The system includes an imaging subsystem that captures low-resolution images of specimens, such as biological or material samples. A computer subsystem processes these images using a CNN with two primary layers. The first layers extract and encode key features from the low-resolution input, creating a compact representation. The second layers reconstruct a high-resolution image from this representation, with the final layer being a sub-pixel convolution layer. This specialized layer refines the output by adjusting pixel-level details, improving sharpness and clarity. The sub-pixel convolution layer is particularly effective in enhancing resolution without introducing artifacts, making it suitable for applications requiring precise image analysis. The system automates the upscaling process, reducing manual intervention and improving efficiency in fields like pathology, materials science, and quality control. The CNN is trained to generalize across different specimen types, ensuring broad applicability.
23. A non-transitory computer-readable medium, storing program instructions executable on one or more computer systems for performing a computer-implemented method for generating a high resolution image for a specimen from a low resolution image of the specimen, wherein the computer-implemented method comprises: acquiring a low resolution image of a specimen; generating a representation of the low resolution image by inputting the low resolution image into one or more first layers of a deep convolutional neural network; and generating a high resolution image for the specimen based on the representation, wherein generating the high resolution image is performed by one or more second layers of the deep convolutional neural network, wherein the one or more second layers comprise a final layer configured to output the high resolution image, wherein the final layer is further configured as a sub-pixel convolution layer, wherein said acquiring, said generating the representation, and said generating the high resolution image are performed by the one or more computer systems, wherein one or more components are executed by the one or more computer systems, and wherein the one or more components comprise the deep convolutional neural network.
This invention relates to image resolution enhancement using deep learning, specifically for generating high-resolution images from low-resolution input images of specimens. The problem addressed is the need for accurate and efficient super-resolution techniques to improve image quality in applications such as microscopy, medical imaging, or remote sensing, where high-resolution data is often unavailable or costly to obtain. The method involves a deep convolutional neural network (CNN) that processes a low-resolution image of a specimen. The low-resolution image is first input into initial layers of the CNN, which generate an intermediate representation of the image. This representation is then passed through subsequent layers, including a final sub-pixel convolution layer, to produce a high-resolution output. The sub-pixel convolution layer is specifically designed to upscale the image by rearranging and interpolating pixel values, ensuring sharpness and detail in the final high-resolution image. The entire process is automated and executed by one or more computer systems, with the CNN being a key component of the system. This approach leverages deep learning to enhance image resolution without requiring multiple low-resolution images or complex optical systems.
24. A computer-implemented method for generating a high resolution image for a specimen from a low resolution image of the specimen, comprising: acquiring a low resolution image of a specimen; generating a representation of the low resolution image by inputting the low resolution image into one or more first layers of a deep convolutional neural network; and generating a high resolution image for the specimen based on the representation, wherein generating the high resolution image is performed by one or more second layers of the deep convolutional neural network, wherein the one or more second layers comprise a final layer configured to output the high resolution image, wherein the final layer is further configured as a sub-pixel convolution layer, wherein said acquiring, said generating the representation, and said generating the high resolution image are performed by one or more computer systems, wherein one or more components are executed by the one or more computer systems, and wherein the one or more components comprise the deep convolutional neural network.
This invention relates to a computer-implemented method for enhancing the resolution of images, specifically for generating high-resolution images from low-resolution input images of specimens. The method addresses the challenge of improving image quality in applications where high-resolution imaging is difficult or impractical, such as in microscopy or medical imaging, where low-resolution images are often the only available data. The method involves acquiring a low-resolution image of a specimen and processing it through a deep convolutional neural network (CNN). The low-resolution image is first input into one or more initial layers of the CNN to generate an intermediate representation. This representation is then processed by additional layers, including a final sub-pixel convolution layer, to produce the high-resolution output. The sub-pixel convolution layer is specifically designed to upscale the image by rearranging and interpolating pixel values, effectively increasing resolution while preserving fine details. The entire process is executed by one or more computer systems, which run the CNN and its components. The neural network is trained to learn the mapping between low-resolution and high-resolution images, enabling accurate reconstruction of high-resolution details from limited input data. This approach leverages deep learning to overcome hardware limitations in imaging systems, providing a cost-effective solution for high-resolution image generation.
Unknown
September 8, 2020
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.